3D Face Modeling and Animation - University Of Illinois · 3D Face Modeling Vuong Le IFP group,...

Post on 26-Sep-2020

1 views 0 download

Transcript of 3D Face Modeling and Animation - University Of Illinois · 3D Face Modeling Vuong Le IFP group,...

3D Face Modeling

Vuong Le IFP group, Beckman Institute University of Illinois ECE417 – Spring 2013

Contents Motivation 3D facial geometry modeling 3D facial geometry acquisition 3D facial deformation modeling Applications Demos

Motivation

Emotive Avatar-based Human-Computer Interaction

Animation Video coding Face recognition, soft biometrics

Face model based video conference

Faceshift [Weise 2011]

3D facial geometry modeling

• Polygonal mesh • 3D morphable model • Implicit surface

Point cloud/Polygonal mesh Consisting Vertices

sample point of the human facial surface can be in correspondence

Triangles Normal vectors Texture

mapped on points/flat triangles

Easy for rendering and low level operations Very high dimensional space for detail models

Candide model [Ahlberg 2001]

3D Morphable Model (1) A human face is not an arbitrary 3D shape Assumption: Every face is built as a combination of known faces

=

a1 * + a2 * + a3 * + a4 * +. . .

b1 * + b2 * + b3 * + b4 * +. . .

Without correspondence

With correspondence

1 __ 2

1 __ 2 + =

3DMM [Vetter 2009]

Average faces across ethnicities

Average faces across nations

3D Morphable Model (2) A vector space of 3D faces Spanned by a PCA basis Learnt from training data

Caricature

Anti Face

Average

Original

Implicit surface Signed distance function volume

- Represented as a function has domain in R3, giving the signed distance to surface - surface as zeros - free space as positive values - occupied space as negative values

- Discretized, limited into voxels of a cube - Can be updated incrementally from pieces of

point cloud - Processing operations can be highly

parallelized - Converted to point cloud by ray tracing or

marching cubes - Example system: KinectFusion, Newcombe

2011

3D facial geometry acquisition

Artist’s designs (skipped) Active acquisition Passive acquisition Reconstruction from 2D images

Active acquisition with Laser radiation Triangulation Laser emitter shines a laser on the subject Camera looks for the location of the laser dot. Depending on how far away the laser strikes a surface, the

laser dot appears at different places in the camera's field of view.

Active acquisition with Infra-red radiation Time of flight Examples: DepthSense , PMD, Creative/Intel perceptual kit

Structured light Example: Kinect, Asus Xtion Pro

Passive Stereo Photogrammetry

Photo courtesy: Dimensional imaging

Beeler - Siggraph 2010

http://www.youtube.com/watch?v=JX5stsU6xfE

3D Face reconstruction from a single 2D image (1) Fitting 3D morphable model

R = Rendering Function To match a face, find optimal α, β, ρ, ι An optimization problem

2α+ ⋅ 3α+ ⋅

3β+ ⋅2β+ ⋅1β ⋅

1α ⋅

,Rρ ι

=

+

+

Iinput

Vetter 2009 Analysis by synthesis strategy

3D World Image

Analysis

Synthesis

Image Model

Image Description

model parameter

3D Face reconstruction from a single 2D image (2) Fast feature point based method

)|,,,,,,( 23 DyxD SfttsP γβα

Neutral Frontal Face

with 2D feature points Model with Texture map

Hu. et. al. FGR2004

Shape model

Generic model

3D facial deformation modeling

Free form Bio-kinetics inspired Data driven methods

Free-form deformation model

The coordinates of the mesh vertices can be deformed in a free-form manner by changing the positions of some control points

Control points can either belong to the mesh vertices or not

Muscle-based deformation model (1)

Facial movement are generated by muscles of the face

Muscle-based deformation model

Build simplified mathematical model that simulates muscle actions on the facial skin

Linear Muscle Models

Facial Action Coding System (FACS) Developed by Ekman and

Friesen, in 1978

FACS describes facial deformations in terms of “Action Units” (AUs) correspond directly to

actions of facial muscles involve to other motions Such as the movement of the tongue or air filling the cheeks

Facial Action Coding System (FACS)

AUs may be combined to describe facial expressions

MPEG4 Facial Animation Parameters (FAPs)

MPEG4 defines 68 FAPs, categorized into 10 groups

Waters, SIGGRAPH 1987

Piecewise Bezier Volume Deformation Model (Tao and Huang, 1998)

Data driven methods - Motion Units

Learn the basic facial deformations from motion capture data

(Hong, Wen, and Huang, 2001)

User specific deformation model

Weise, Siggraph 2011

Applications

Face recognition Face tracking Expression recognition (Emotion prediction) Talking avatar

Face recognition

Face tracking and Expression recognition

3D facial animation - Avatar Key-frame interpolation method Place particular facial deformations at particular time instants

(key-frames) Facial deformations in-between key-frames are obtained by a

certain interpolation scheme

Two basic key-frame types Visemes Expressions

Phonemes and Visemes

A Phoneme: basic acoustic unit of sound A Viseme: Visual counterpart of a phoneme

Weise, Li – Faceshift - Siggraph 2011

Performance driven avatar

Text-driven talking heads

Speech-driven talking heads

Preview of MP5

End of main slides

CANDIDE-3 model refined model face scan model fit to scan mapped